Hidden Markov Model-based Formulations of Sensor Scheduling in Dynamic Environments

Real world surveillance applications include heterogeneous sensors, which trade off performance (e.g., detection, identification, and tracking accuracies) versus the sensor usage cost (e.g., power and bandwidth consumption, distance traveled, risk of exposure, deployment requirements). The objective of sensor scheduling is to judiciously allocate sensing resources among competing tasks dynamically to exploit the individual sensor capabilities to accurately estimate task states, while minimizing the sensor usage cost. A HMM with controllable emission matrices corresponding to each sensor is an appropriate way to model the sensor scheduling problem, because the task\u27s activities are partially observed, and their true states can only be inferred through uncertain observations obtained by the sensing assets. Motivated by the intelligence, surveillance, and reconnaissance (ISR) and Anti-Submarine Warfare (ASW) search applications, a series of sensor scheduling problems ranging from a single HMM to multiple coupled HMMs are formulated and near-optimal scheduling strategies are proposed to overcome the computational intractability of the dynamic programming (DP) recursion associated with the optimal policy.